Snowflake and Databricks bank PostgreSQL acquisitions to bring transactions onto their platforms

In the past few weeks, both Snowflake and Databricks dipped into their respective acquisition funds and found the resources to buy specialist providers of PostgreSQL database systems. In May, Databricks, a data lake maker originally built around Apache Spark, bought Neon for a reported $1 billion in equity. Neon, which builds …

[Gordon 10]

The Gartner write up is bollocks

OLTP and Analytics DB's are a false seperation for all but the biggest workloads.

90% of Postgres databases will fit into Databricks / Snowflake with very little problem because the cloud DB's have the horsepower to brute force the performance.

Its solely inertia and cost of change that persists Postgres at this point. The DB's are probably 2nd tier apps that are chugging along for years without change.

[uberbloke]

Re: The Gartner write up is bollocks

Not quite.

Databricks is only ACID compliant at the statement level (no support for transactions spanning multiple statements) and primary and foreign keys are informational and not enforced, so for OLTP, regardless of size they are not really suitable, and even for analytics, try telling your finance head that the reporting system doesn't, natively, ensure that duplicate keys or orphans are excluded

Hopefully this is the sort of thing that these acquisitions will address.

[logicalextreme]

Re: The Gartner write up is bollocks

Same with Snowflake. It's like using old MySQL with the default storage engines — it'll happily let you declare all sorts of essential constraints and then promptly (and silently) ignore them. Cue reams more code if you want to do even the most basic checks on incoming data.

They do offer something called "hybrid tables" which are essentially rowstore rather than columnar and offer a smattering of RI, but they're not really the point of Snowflake AFAICT and they're mysteriously not available in Azure. Like, at all. ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯